Cross Media Attribution

Building bridges across Paid, Owned and Earned Media
to optimize media mix efficiencies and effectiveness
Cross Media
Attribution
May 2012

© 2012 Havas Media Group 1
Cross Media AttributionHAVAS MEDIA GROUP INSIGHT
Lead Contributors
Katrin Ribant
Data Platforms
Havas Media Group
Dejan Duzevik
Chief Technology Officer
Concentric
dejan@concentricroi.com
Arnaud Parent
Director, Research & Development, PhD
Havas Media
arnaud.parent@@havasmedia.com
Sylvain Le Borgne
Insights & Data Solutions
Artemis
sylvain.leborgne@havasmg.com
Doug Watson
Data Solutions
Artemis
doug.watson@havasmg.com

Increased fragmentation of the media space and constant innovation in the
digital media space pose a challenge to Marketers who need to understand how
to optimally combine communication vehicles to reach their marketing goals.
At the center of this challenge is the individual consumer who is confronted
with an extreme number of choices in brands, products, services and media
messaging. Today’s consumer behaviors and choices are getting more complex
and harder to understand. These complexities have rendered some of the tradi-
tional tools that Marketers use for media mix optimization less efficient.
Boundaries between traditional and digital media are fading, as are limits be-
tween Paid, Owned and Earned marketing actions. Silos make less sense than
ever. A holistic approach that considers the overall marketing strategy is needed
to help Marketers achieve their goals efficiently and effectively.
Executive summary
Havas Media Group is proposing new approaches to help Mar-
keters understand how different combinations of messaging ve-
hicles can help them achieve their goals, leveraging synergies be-
tween channels using a 100% fact-based, data-driven approach.
Artemis Attribution for Paid, Owned and Earned Media helps
Marketers optimize spend whilst taking into account how differ-
ent channels impact different consumer segments in real-time.
Based on simulation methods born in the science of complexity, this new frame-
work for attribution management builds on algorithmic digital attribution and
has the capability to integrate results from econometric modeling to help Mar-
keters understand how their efforts can help get their message across to the
right people at the right time.
As we continue developing our toolset, we will publish more results in further Insights documents.
In the meantime, we would be delighted to discuss individual business issues and opportunities,
whether you are already a Havas Media Group client or simply just interested in a more case-
specific perspective. Please contact your Havas Media Group Account Director, the appropriate
Havas Media Group country Managing Director or any authors to find out how the emerging op-
portunities in digital marketing can play a key role in your overall marketing plans.

Ad serving, social and web analytics data enable Marketers to have access
to more extended and precise insights on consumer behavior. This wealth of
information means that differences between consumer segments are better un-
derstood today than ever before.
These differences are instrumental in designing effective and efficient cam-
paign architectures that drive the right message to the right people. Traditional
marketing mix modeling methodologies, like econometrics, are not designed to
easily take these differences into account.
Increased complexity of the media space means Marketers need to optimize
campaigns that have multiple objectives and target groups. Moreover, the in-
creasing complexity that Marketers have to handle when measuring results of
multi-channel campaigns and business results means that there is not one an-
swer that fits a given set of circumstances. Rather, Marketers need the flexibility
to experiment with different scenarios and understand how a change of inputs
influences different consumer segments (i.e. running TV and paid search to-
gether increases web base sales for product A).
Why a new approach for performance
attribution?
In this Insights piece, we look at 3 different client cases
involving the following:
• A CPG advertiser who wants to know if investing in their Facebook
presence can help to increase consumer engagement to eventually
increase sales.
• An ecommerce advertiser using social media, including Twitter as a
Customer Relationship Management channel. This Marketer wants to
maximize his most profitable customer segment.
• A finance advertiser using TV, search and display as well as social to
drive online applications for insurance subscriptions. This Marketer’s
goal is to increase valid online applications.

We do not believe that there is one answer to any Marketer’s attribution ques-
tions. Therefore, we focus on enabling Marketers to experiment with “what if”
scenarios allowing them to understand how different mixes of media channels,
social connectivity, competitive offering and other contextual influences result
in different outcomes.
Enabling “what if” scenarios
To enable relevant “what if” scenario building, we create an ecosystem which
mirrors a given historical period of time. This is done in 4 steps:
How is Havas Media Group’s
approach different?
What
if?
Shift 3%
from TV to
search
Shift 5%
to dis-
play
Modify
campaign
flights
Allocate 45%
to brand
campaigns
Phase TV to
search
step 1 step 2 step 3 step 4
Integrate data sets Create ecosystem
Build multi-dimensional
attribution model
Experiment “what if”
scenarios
User data E-mail Twitter Facebook TV
RadioSocial Sites Analytics Search Display

Step 1: Integrate data sets
We gather data relative to the Marketer’s activity. This ranges from digital/of-
fline campaign data to web analytics or social data, as well as data about the
Marketer’s competition and product offerings.
After a thorough analysis of business needs and available data sources, data is
selected and gathered according to how relevant it is to answer the Marketer’s
questions. This step is about identifying “useful big data” rather then simply as-
sembling all available data sets.
Step 2: Create ecosystem
Using Agent Based Modeling, a leading-edge, mathematical simulation meth-
odology, we create an ecosystem that will reproduce the outputs of the data
we integrated.
This ecosystem is based on a series of simulated users (the more the better, as
shown in the case studies in this paper, we often work with 1 Billion+ users) that
act as consumers, using media, social buzz and connectivity data to determine
their purchase behavior.
step 1
step 2

To check if the ecosystem gives credible results, we will calibrate it against histori-
cal data. This means that an ecosystem should reproduce real historical results
with a high level of accuracy before we can use it to build the “what if” scenarios.
Social Network
with lower density
Recommendation: This customer
segment is better activated with
media that generates high awareness.
Social Network
with higher density
Recommendation: This customer
segment is better activated with
media that generates word of mouth.
Analysis: Awareness of product
or promotion propagates slower
Analysis: Awareness of product
or promotion propagates quicker
What is Agent Based Modeling?
Agent Based Modeling (ABM) aggregates disparate data to create “agents”
(simulated people) that are representative of the real world population.
ABMs are used to uncover the causes of complex and often surprising out-
comes in systems as diverse as molecules, the Internet, traffic jams and,
now, online consumer journeys.
Because ABMs are designed to understand individuals, they are suited to
help Marketers understand how different user segments behave as a func-
tion of marketing campaign exposure. For more on ABMs see:
http://en.wikipedia.org/wiki/Agent-based_model
Here is an example of what a social network looks like for 2 different clusters
of users. The right cluster is clearly the denser of the two. This means that
information will propagate a lot faster in this cluster and, for example, influ-
ence how awareness of a product offering is increased among users.
GEEK’S
CORNER

Step 4: Experiment with “what if” scenarios
The purpose of creating “what if” scenarios is to answer questions about media
mix optimization. These questions could be:
• What is an optimal allocation between online and offline?
• What is the right allocation between search and display?
• What is the value of having a growing fan base?
• Does my connectivity in the social sphere positively influence sales?
step 3
What if?
Shift 3% from
TV to search
Shift 5% to display Modify campaign flights
Allocate 45% to
brand campaigns
Phase TV to search
step 4
Artemis Attribution for Paid, Owned and Earned Media is fully transparent,
evidence-based and data driven. We take all media vehicles into account in a
truly cross-channel approach that addresses the complexity and diversity of
today’s media landscapes and consumer behaviors.
Step 3: Build multi-dimensional attribution model
The system outputs an attribution model according to the role each influence
plays in the consumer’s journey to decision making. This model is fully trans-
parent and can be tweaked and calibrated using the Marketers business knowl-
edge to connect potential skews.

This case study looks at the CPG category. This CPG Marketer runs a loyalty
program on Facebook for one of their flagship products. The program focuses
on healthy food and lifestyle recommendations and is aimed at creating a moti-
vational community fostering better and healthier lifestyle habits.
Many Marketers struggle to understand how to properly balance investments
between new users and advocates. Using Artemis Attribution to analyze this
CPG Marketer’s Paid, Owned and Earned Media activities, we provided an inte-
grated view of performance and optimized their dual strategies.
Marketer’s concerns
• Digital budget setting and how to allocate spend between channels,
specifically share of spend between video and social.
• How to articulate digital campaigns with TV campaigns to:
— Maximize synergies with TV spend.
— Understand how to use channels to leverage reach on different
consumer segments as well as new/loyal customers.
KPIs
We selected a variety of online KPIs that are considered as indicators of en-
gagement and predictive of the consumer’s purchase behavior such as sub-
scription to the Marketer’s loyalty program, time spent/frequency of visits on
the program’s micro site and coupon downloads.
CASE 1
CPG MarketER
Impact of fan base growth on sales
Artemis Attribution for POE Media allows Marketers to create as many of these
scenarios as they wish to explore and compare effects of different strategies
on overall sales (or other variables defined as outcomes). Continuous injection
of real-time data informs allocation and re-allocation of spend according to real-
world results of these decisions. This allows for real-time piloting of media strate-
gies adapted to a rapidly changing context. In the following section, we will out-
line three use cases showing how we answered some of our client’s questions.

Strategy
The advertising program runs display, video and search, while promoting the
Marketer’s Facebook page and driving people to a reduction coupon on their
website.
Channels used
Video Display Search Facebook
pages

Digital Mix: Majority of volume is concentrated on video and display
Video and display dominate
the digital mix. Social ads
campaign is minimal.
Loyalty program has been
successful on Facebook,
with the help of display,
video and search ads.
Video and display clicks vs search clicks
The following section details our approach to:
a) Attribution: Last click versus Artemis Attribution for POE Media
b) Media Mix Scenarios: Shifting budget from video to display or
invest in building fan base?
The following chart shows how digital spend is concentrated on video and dis-
play. Search accounts for 15% of clicks for this marketing program. This media
mix helped the Marketer build a sizeable fan base of over 300,000 fans.

Attribution: Last click attribution favors video and does not credit social
Artemis POE Attribution takes
credit away from video
Display and the Marketer’s social
presence on Facebook gain credit
Fan Base
Artemis POE
Attribution 10%
-22% attribution
+12% attribution
Last click attribution favors video, does not credit social
The re-allocation of credit from video to display makes sense as display usually
plays a role in earlier stages of the funnel. Display and video have complementary
roles. Video takes over where display appears, diminishing returns and acting as a
complementary channel for TV. This exposes the Marketer’s message to consum-
er segments that are difficult to reach through traditional TV actions and engages
all consumer segments with a direct link to the Marketer’s loyalty program.
Since coupon downloads from Facebook fans cannot be tracked directly, we ran
Artemis Attribution for Paid, Owned and Earned Media and determined that
10% of conversions could be credited to the advertiser’s fan base.
a) Attribution: Last click versus Artemis Attribution
for POE Media
Video (the darkest blue bar on the chart below) takes the majority of attribu-
tion based on the last click model. Despite the relative balance of display, video
impressions and clicks (see pie charts), it seems video has a greater impact
than display.

Based on the re-allocation of conversions credited to display and social by Ar-
temis POE Attribution, we explored a scenario of media spend optimization and
proposed a move of 10% of spend from video to display.
Move 10% of Spend from video to display
b) Media mix scenarios: Shifting budget from
video to display or invest in building fan base?
Based on the re-allocation of conversions credited to display and social by Ar-
temis POE Attribution, we explored two scenarios of media spend optimization:
• Move 10% of spend from video to dispaly
• Invest in building fan base
Drop in conversions
if spend shifts
to display
Conversion trend is
constant but volume
drops if spend shifts
to display
Shifting 10% of spend to display would result in less coupon downloads
Shift in spend to display causes 8% drop in coupon downloads
Drop in conversion
Drop in conversion

The more the fan base increases, the higher the projected growth in conversions.
If the Marketer were to double the size of their fan base, conversions would in-
crease by 23%. We recommend investing in social ad campaigns and specific
social actions that will help grow this Marketer’s Facebook presence. This will
foster engagement in their loyalty program and lead consumers to keep the
Marketer’s product top of mind, while resulting in higher sales volumes.
We explored what would happen if the Marketer was to increase their fan base
and created two scenarios based on our findings:
• Increasing fans by 20%
• Increasing fans by 100%
For this advertiser, investing in growing their social presence results in an in-
crease in conversions. Since the display channel is saturated and Artemis At-
tribution for POE Media credits the Marketer’s social presence for 10% of con-
versions, we discovered that growing their fan base can help them increase
conversion volumes in the short term and sales in the long term.
Invest in building fan base
Despite the fact that more credit was attributed to display by Artemis Attribu-
tion for POE Media, this scenario predicts an 8% drop in coupon downloads.
The explanation for this drop is that the display channel has reached its level of
diminishing returns for this Marketer. Additional spend in display will not result
in better performance. Therefore, we need to explore alternative channels.
Conversions increase with fans
Increasing fan base increases conversions
Doubling fans increases
conversions by 23%

Internet is instrumental to this ecommerce advertiser’s media mix and sales.
Although their business model was traditionally bricks and mortar, they suc-
cessfully shifted to the web as a distribution channel. Today, about 60% of sales
happen on their website. Their products call for a purchase window from a few
days to several weeks and peer reviews are important to consumer’s choice.
This Marketer’s goal is to understand:
• What digital media mix will result in more customers
and more of their most valuable customers?
• Can they impact sales through the use of social media?
To answer these questions, we analyzed:
• 12 months of data
• 100+ campaign
• 1 billion user journeys
CASE 2
eCommerce Marketer
Impacting sales through social media
• 33 billion events
• Digital and social media channels
Brick and Mortar
Sales 40%
Online Sales 60%
60% of sales happen on the Marketer’s website

Channels
used
Display Search Price
Comparison
Affiliates E-mail Twitter
Question 1: Optimizing the digital media mix
First, we looked at channel attribution to answer the Marketer’s question: What
digital media mix will result in more customers and more of my most valuable
customers?
a) Channel impact
We determined that credit for sales that were previously accounted to paid
search was shifting to display. In fact, attribution to display more than doubled.
This clearly indicates that display is helping customers make their choice at ear-
lier stages than search.
We found more
evidence of the
impact of display
at earlier stages of
the decision mak-
ing process by
running a channel
impact analysis.
Channel attribution shifts credit to display
15% reallocation to display
Display’s
attribution
more than 2x
Attribution reallocates 15% credit from search to display
Last Click attribution by Channel Attributed conversions by Channel

b) Customer value
To determine customer value, we segmented customers into three categories
according to their revenue/purchase ratio.
As shown on the pie chart on the following page, about half of the revenue
comes from Platinum customers (darkest blue), the top 20% most valuable cus-
tomers. So the 20/80 rule becomes the 20/50 rule for this Marketer.
Platinum customers are six times more valuable than Silver customers and three
times more valuable than an average customer as shown on the bar/line chart
on the following page. This means that focusing on attracting an additional
Platinum customer should bring three times more revenue than acquiring an av-
erage customer. Therefore, this Marketer’s strategy should focus on analysis of
parameters that grow the acquisition and lower the levels of churn of Platinum
customers as a priority.
The left part of the chart below represents the moment of purchase. Both search
(lighter blue line) and display (darker blue line) have the greatest impact close
to the moment of purchase, with search being more impactful than display.
Display has a greater impact than search 6 to 14 days prior to purchase (the
darker blue line is higher than the lighter blue line from 6 to 14 days prior to
purchase).
Display has a greater impact than search at early stages of the purchase funnel
Display and search have maximal
impact close to purchase
6 to 14 days prior to purchase, display
has a higher impact than search
Impact of display and search up to 14 days before purchase
Purchase 1 week before purchase 2 weeks before purchase

c) “What if” scenarios:
Optimizing for the most valuable customers
In order to maximize overall sales volume with a focus on maximizing Platinum
customer acquisition, we analyzed two scenarios for moving spend from search
to display. First, moving 5% of spend to display and then moving 10% of spend
to display.
Both scenarios resulted in increased sales. Moving 5% of spend to display is the
most efficient scenario, both for overall sales volume increase and for Platinum
clients volume increase. Moving 10% of spend to display increases overall sales
volume less than moving 5% of spend to display, resulting in acquisition of less
valuable customers.
Platinum customers 6x more
valuable than Silver customers
Platinum customer
account for 50%
of revenue
Platinum customers 6x more valuable then Silver, 3x vs average
Revenue by customer segment
20% of customers account for 50% of the revenue

Question 2: Can social media impact sales?
First, we looked at channel attribution to answer the Marketer’s question: What
digital media mix will result in more customers and more of my most valuable
customers?
Moving 5% of search
spend to display results
in increased sales
Proportion of Platinum
customers increases if 5% of
spend is moved to display
What If Scenarios: impact on customer segments (absolute values)
What If Scenarios: impact on customer segments (percentages)
Moving 5% of search spend to display brings best results

In this graph, we compare sales for the same time pe-
riod as above but have projected a scenario in which
the Marketer has higher connectivity than the industry
average. Not only is the recovery faster but also most
seasonal peaks are amplified by the spread of the Mar-
keter’s promotion through online word of mouth.
The first chart looks at a projection of sales using the av-
erage connectivity of the Marketer’s space. We project
what the recovery (dark blue line) would have looked
like if the Marketer’s connectivity in the social space was
on par with the average for their industry. If the Market-
er had average connectivity in the social space, the dip
in sales would have been less and the recovery would
have been faster.
Sales projection if Marketer had average connectivity in the social space.
Sales projection if Marketer had higher than average connectivity in the social space.
High social connectivity
a) Connectivity in the social space influences sales
During the concerned time period, a news event negatively impacted this Mar-
keter’s category. After this event, sales dropped over several weeks. We built
scenarios examining how spend allocation and efforts in social media could
have helped sales grow back to normal quicker.
Average social connectivity
seasonal peaks
are higher
recovery is
faster
higher connectivity helps message
spread and increase sales

Valid insurance
subscription
In our third case, we look at a Finance brand whose sales process involves driv-
ing people to their website where they can apply for an insurance policy. To
drive people to apply online, our Finance Marketer uses a mix of digital and of-
fline channels including paid search and TV.
Although this Marketer’s media mix is far more complex and makes use of mul-
tiple channels, including display, social and radio, we chose to focus on one par-
ticular campaign and narrow down to TV and search only. This has been done in
order to better assess the Marketer’s questions about digital and offline phasing.
TV’s main role for this Marketer is to raise awareness and drive online search for
specific keywords. Search engine marketing is directly aimed at views on the
Marketer’s website where people consider applying for an insurance policy.
Once the prospect posts their insurance policy application online, the lead gen-
erated goes through a validation process, resulting in a data passback between
the Marketer’s data systems and Artemis.
Channels used
by Finance brand
CASE 3
FINANCE MARKETER
Phasing TV and search to drive subscriptions
TV Search
Online
application
Validation
As Artemis Attribution maps the effect of Paid Media on Earned and Owned
Media, it helps Marketers who are faced with unforeseeable events affecting
their sales. It assists with the design of Paid Media actions that help boost their
social presence in the recovery phase of such an event.

Since shifting 5% of spend results
in a higher increase in valid appli-
cations, we recommend up weight-
ing the paid search campaign and
reassessing performance during
the following month.
However, shifting spend from TV to
search in order to improve results
needs to be implemented carefully.
Paid search represents 8% of TV spend
Moving spend to search increases valid applications
Baseline
The 2 scenarios we explored involve shifting spend from TV to paid search in
order to increase valid applications:
• Shifting 3% of spend to paid search adding $210k to current $567k spend
• Shifting 5% of spend to paid search adding $350k to current $567k spend
Both scenarios result in an increase of valid applications.
The Marketer’s goal is to drive more succesful applications, meaning more ap-
plications that actually result in an insurance subscription.
KPIs
Online applications validated through passback of completion data from the
Marketer’s database.

Paid search conversion rates are 14% higher on average when a TV campaign is
aired. Therefore, we recommend that additional budgets are concentrated on
the time periods where TV is active.
Adapting phasing of additional search budgets in real-time according to how
they are impacted by TV actions could help boost this Marketer’s KPIs even
further by maximizing digital and offline synergies. In further stages of opti-
mization, we will broaden the scope of the analysis to include social and other
paid channels.
Search conversion rate is 14% higher when TV is on air

In the light of the ever growing complexity of the digital media landscape, few
Marketers still believe that there is one perfect answer to their spend allocation
challenges. If there were to be such a solution, it would have a short lifespan
and quickly be disrupted by a shift in consumer behavior fostered by the digital
media revolution.
Havas Media Group has made certain the continuous development of tools and
methodologies that help Marketers manage and optimize their efforts, in accord-
ance with the level of complexity of their vertical and consumer segments. We do
not believe in one size fits all marketing strategies nor relationship management.
As we continue developing our toolset, we will publish more results in further
Insights documents. In the meantime, we would be delighted to discuss individ-
ual business issues and opportunities with you, whether you are already a Havas
Media Group client or simply just interested in a more case-specific perspec-
tive. Please contact your Havas Media Group Account Director, the appropriate
Havas Media Group country Managing Director or any authors of this Insight to
find out how many of the emerging opportunities in digital marketing can play
a key role in your overall marketing goals.
Conclusion

www.havasmediagroup.com
Produced by Havas Media Group

Cross Media Attribution

Recommended

Recommended

More Related Content

What's hot

What's hot (20)

Viewers also liked

Viewers also liked (13)

Similar to Cross Media Attribution

Similar to Cross Media Attribution (20)

Recently uploaded

Recently uploaded (20)

Cross Media Attribution